1 /*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
5 * Copyright (C) 2009 VMware, Inc. All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
23 * OTHER DEALINGS IN THE SOFTWARE.
24 */
25
26
27 /**
28 * \file tnl/t_vb_program.c
29 * \brief Pipeline stage for executing vertex programs.
30 * \author Brian Paul, Keith Whitwell
31 */
32
33
34 #include "main/glheader.h"
35 #include "main/macros.h"
36 #include "main/samplerobj.h"
37 #include "main/state.h"
38 #include "math/m_xform.h"
39 #include "program/prog_instruction.h"
40 #include "program/prog_statevars.h"
41 #include "program/prog_execute.h"
42 #include "swrast/s_context.h"
43 #include "util/bitscan.h"
44 #include "util/u_memory.h"
45
46 #include "tnl/tnl.h"
47 #include "tnl/t_context.h"
48 #include "tnl/t_pipeline.h"
49
50
51 #ifdef NAN_CHECK
52 /** Check for NaNs and very large values */
53 static inline void
check_float(float x)54 check_float(float x)
55 {
56 assert(!util_is_inf_or_nan(x));
57 assert(1.0e-15 <= x && x <= 1.0e15);
58 }
59 #endif
60
61
62 /*!
63 * Private storage for the vertex program pipeline stage.
64 */
65 struct vp_stage_data {
66 /** The results of running the vertex program go into these arrays. */
67 GLvector4f results[VARYING_SLOT_MAX];
68
69 GLvector4f ndcCoords; /**< normalized device coords */
70 GLubyte *clipmask; /**< clip flags */
71 GLubyte ormask, andmask; /**< for clipping */
72
73 GLboolean vertex_textures;
74
75 struct gl_program_machine machine;
76 };
77
78
79 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
80
81
82 static void
userclip(struct gl_context * ctx,GLvector4f * clip,GLubyte * clipmask,GLubyte * clipormask,GLubyte * clipandmask)83 userclip( struct gl_context *ctx,
84 GLvector4f *clip,
85 GLubyte *clipmask,
86 GLubyte *clipormask,
87 GLubyte *clipandmask )
88 {
89 GLbitfield mask = ctx->Transform.ClipPlanesEnabled;
90 while (mask) {
91 const int p = u_bit_scan(&mask);
92 GLuint nr, i;
93 const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
94 const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
95 const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
96 const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
97 GLfloat *coord = (GLfloat *)clip->data;
98 GLuint stride = clip->stride;
99 GLuint count = clip->count;
100
101 for (nr = 0, i = 0 ; i < count ; i++) {
102 GLfloat dp = (coord[0] * a +
103 coord[1] * b +
104 coord[2] * c +
105 coord[3] * d);
106
107 if (dp < 0) {
108 nr++;
109 clipmask[i] |= CLIP_USER_BIT;
110 }
111
112 STRIDE_F(coord, stride);
113 }
114
115 if (nr > 0) {
116 *clipormask |= CLIP_USER_BIT;
117 if (nr == count) {
118 *clipandmask |= CLIP_USER_BIT;
119 return;
120 }
121 }
122 }
123 }
124
125
126 static GLboolean
do_ndc_cliptest(struct gl_context * ctx,struct vp_stage_data * store)127 do_ndc_cliptest(struct gl_context *ctx, struct vp_stage_data *store)
128 {
129 TNLcontext *tnl = TNL_CONTEXT(ctx);
130 struct vertex_buffer *VB = &tnl->vb;
131 /* Cliptest and perspective divide. Clip functions must clear
132 * the clipmask.
133 */
134 store->ormask = 0;
135 store->andmask = CLIP_FRUSTUM_BITS;
136
137 tnl_clip_prepare(ctx);
138
139 if (tnl->NeedNdcCoords) {
140 VB->NdcPtr =
141 _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
142 &store->ndcCoords,
143 store->clipmask,
144 &store->ormask,
145 &store->andmask,
146 !(ctx->Transform.DepthClampNear &&
147 ctx->Transform.DepthClampFar) );
148 }
149 else {
150 VB->NdcPtr = NULL;
151 _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
152 NULL,
153 store->clipmask,
154 &store->ormask,
155 &store->andmask,
156 !(ctx->Transform.DepthClampNear &&
157 ctx->Transform.DepthClampFar) );
158 }
159
160 if (store->andmask) {
161 /* All vertices are outside the frustum */
162 return GL_FALSE;
163 }
164
165 /* Test userclip planes. This contributes to VB->ClipMask.
166 */
167 /** XXX NEW_SLANG _Enabled ??? */
168 if (ctx->Transform.ClipPlanesEnabled &&
169 (!_mesa_arb_vertex_program_enabled(ctx) ||
170 ctx->VertexProgram.Current->arb.IsPositionInvariant)) {
171 userclip( ctx,
172 VB->ClipPtr,
173 store->clipmask,
174 &store->ormask,
175 &store->andmask );
176
177 if (store->andmask) {
178 return GL_FALSE;
179 }
180 }
181
182 VB->ClipAndMask = store->andmask;
183 VB->ClipOrMask = store->ormask;
184 VB->ClipMask = store->clipmask;
185
186 return GL_TRUE;
187 }
188
189
190 /**
191 * XXX the texture sampling code in this module is a bit of a hack.
192 * The texture sampling code is in swrast, though it doesn't have any
193 * real dependencies on the rest of swrast. It should probably be
194 * moved into main/ someday.
195 */
196 static void
vp_fetch_texel(struct gl_context * ctx,const GLfloat texcoord[4],GLfloat lambda,GLuint unit,GLfloat color[4])197 vp_fetch_texel(struct gl_context *ctx, const GLfloat texcoord[4], GLfloat lambda,
198 GLuint unit, GLfloat color[4])
199 {
200 SWcontext *swrast = SWRAST_CONTEXT(ctx);
201
202 /* XXX use a float-valued TextureSample routine here!!! */
203 swrast->TextureSample[unit](ctx, _mesa_get_samplerobj(ctx, unit),
204 ctx->Texture.Unit[unit]._Current,
205 1, (const GLfloat (*)[4]) texcoord,
206 &lambda, (GLfloat (*)[4]) color);
207 }
208
209
210 /**
211 * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
212 * string has been parsed.
213 */
214 GLboolean
_tnl_program_string(struct gl_context * ctx,GLenum target,struct gl_program * program)215 _tnl_program_string(struct gl_context *ctx, GLenum target, struct gl_program *program)
216 {
217 /* No-op.
218 * If we had derived anything from the program that was private to this
219 * stage we'd recompute/validate it here.
220 */
221 return GL_TRUE;
222 }
223
224
225 /**
226 * Initialize virtual machine state prior to executing vertex program.
227 */
228 static void
init_machine(struct gl_context * ctx,struct gl_program_machine * machine,GLuint instID)229 init_machine(struct gl_context *ctx, struct gl_program_machine *machine,
230 GLuint instID)
231 {
232 /* Input registers get initialized from the current vertex attribs */
233 memcpy(machine->VertAttribs, ctx->Current.Attrib,
234 MAX_VERTEX_GENERIC_ATTRIBS * 4 * sizeof(GLfloat));
235
236 machine->NumDeriv = 0;
237
238 /* init call stack */
239 machine->StackDepth = 0;
240
241 machine->FetchTexelLod = vp_fetch_texel;
242 machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
243
244 machine->Samplers = ctx->VertexProgram._Current->SamplerUnits;
245
246 machine->SystemValues[SYSTEM_VALUE_INSTANCE_ID][0] = (GLfloat) instID;
247 }
248
249
250 /**
251 * Map the texture images which the vertex program will access (if any).
252 */
253 static void
map_textures(struct gl_context * ctx,const struct gl_program * vp)254 map_textures(struct gl_context *ctx, const struct gl_program *vp)
255 {
256 GLuint u;
257
258 for (u = 0; u < ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits; u++) {
259 if (vp->TexturesUsed[u]) {
260 /* Note: _Current *should* correspond to the target indicated
261 * in TexturesUsed[u].
262 */
263 _swrast_map_texture(ctx, ctx->Texture.Unit[u]._Current);
264 }
265 }
266 }
267
268
269 /**
270 * Unmap the texture images which were used by the vertex program (if any).
271 */
272 static void
unmap_textures(struct gl_context * ctx,const struct gl_program * vp)273 unmap_textures(struct gl_context *ctx, const struct gl_program *vp)
274 {
275 GLuint u;
276
277 for (u = 0; u < ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits; u++) {
278 if (vp->TexturesUsed[u]) {
279 /* Note: _Current *should* correspond to the target indicated
280 * in TexturesUsed[u].
281 */
282 _swrast_unmap_texture(ctx, ctx->Texture.Unit[u]._Current);
283 }
284 }
285 }
286
287
288 /**
289 * This function executes vertex programs
290 */
291 static GLboolean
run_vp(struct gl_context * ctx,struct tnl_pipeline_stage * stage)292 run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
293 {
294 TNLcontext *tnl = TNL_CONTEXT(ctx);
295 struct vp_stage_data *store = VP_STAGE_DATA(stage);
296 struct vertex_buffer *VB = &tnl->vb;
297 struct gl_program *program = ctx->VertexProgram._Current;
298 struct gl_program_machine *machine = &store->machine;
299 GLuint outputs[VARYING_SLOT_MAX], numOutputs;
300 GLuint i, j;
301
302 if (!program)
303 return GL_TRUE;
304
305 /* ARB program or vertex shader */
306 _mesa_load_state_parameters(ctx, program->Parameters);
307
308 /* make list of outputs to save some time below */
309 numOutputs = 0;
310 for (i = 0; i < VARYING_SLOT_MAX; i++) {
311 if (program->info.outputs_written & BITFIELD64_BIT(i)) {
312 outputs[numOutputs++] = i;
313 }
314 }
315
316 /* Allocate result vectors. We delay this until now to avoid allocating
317 * memory that would never be used if we don't run the software tnl pipeline.
318 */
319 if (!store->results[0].storage) {
320 for (i = 0; i < VARYING_SLOT_MAX; i++) {
321 assert(!store->results[i].storage);
322 _mesa_vector4f_alloc( &store->results[i], 0, VB->Size, 32 );
323 store->results[i].size = 4;
324 }
325 }
326
327 map_textures(ctx, program);
328
329 for (i = 0; i < VB->Count; i++) {
330 GLuint attr;
331
332 init_machine(ctx, machine, tnl->CurInstance);
333
334 #if 0
335 printf("Input %d: %f, %f, %f, %f\n", i,
336 VB->AttribPtr[0]->data[i][0],
337 VB->AttribPtr[0]->data[i][1],
338 VB->AttribPtr[0]->data[i][2],
339 VB->AttribPtr[0]->data[i][3]);
340 printf(" color: %f, %f, %f, %f\n",
341 VB->AttribPtr[3]->data[i][0],
342 VB->AttribPtr[3]->data[i][1],
343 VB->AttribPtr[3]->data[i][2],
344 VB->AttribPtr[3]->data[i][3]);
345 printf(" normal: %f, %f, %f, %f\n",
346 VB->AttribPtr[2]->data[i][0],
347 VB->AttribPtr[2]->data[i][1],
348 VB->AttribPtr[2]->data[i][2],
349 VB->AttribPtr[2]->data[i][3]);
350 #endif
351
352 /* the vertex array case */
353 for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
354 if (program->info.inputs_read & BITFIELD64_BIT(attr)) {
355 const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
356 const GLuint size = VB->AttribPtr[attr]->size;
357 const GLuint stride = VB->AttribPtr[attr]->stride;
358 const GLfloat *data = (GLfloat *) (ptr + stride * i);
359 #ifdef NAN_CHECK
360 check_float(data[0]);
361 check_float(data[1]);
362 check_float(data[2]);
363 check_float(data[3]);
364 #endif
365 COPY_CLEAN_4V(machine->VertAttribs[attr], size, data);
366 }
367 }
368
369 /* execute the program */
370 _mesa_execute_program(ctx, program, machine);
371
372 /* copy the output registers into the VB->attribs arrays */
373 for (j = 0; j < numOutputs; j++) {
374 const GLuint attr = outputs[j];
375 #ifdef NAN_CHECK
376 check_float(machine->Outputs[attr][0]);
377 check_float(machine->Outputs[attr][1]);
378 check_float(machine->Outputs[attr][2]);
379 check_float(machine->Outputs[attr][3]);
380 #endif
381 COPY_4V(store->results[attr].data[i], machine->Outputs[attr]);
382 }
383
384 /* FOGC is a special case. Fragment shader expects (f,0,0,1) */
385 if (program->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_FOGC)) {
386 store->results[VARYING_SLOT_FOGC].data[i][1] = 0.0;
387 store->results[VARYING_SLOT_FOGC].data[i][2] = 0.0;
388 store->results[VARYING_SLOT_FOGC].data[i][3] = 1.0;
389 }
390 #ifdef NAN_CHECK
391 assert(machine->Outputs[0][3] != 0.0F);
392 #endif
393 #if 0
394 printf("HPOS: %f %f %f %f\n",
395 machine->Outputs[0][0],
396 machine->Outputs[0][1],
397 machine->Outputs[0][2],
398 machine->Outputs[0][3]);
399 #endif
400 }
401
402 unmap_textures(ctx, program);
403
404 if (program->arb.IsPositionInvariant) {
405 /* We need the exact same transform as in the fixed function path here
406 * to guarantee invariance, depending on compiler optimization flags
407 * results could be different otherwise.
408 */
409 VB->ClipPtr = TransformRaw( &store->results[0],
410 &ctx->_ModelProjectMatrix,
411 VB->AttribPtr[0] );
412
413 /* Drivers expect this to be clean to element 4...
414 */
415 switch (VB->ClipPtr->size) {
416 case 1:
417 /* impossible */
418 case 2:
419 _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
420 /* fall-through */
421 case 3:
422 _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
423 /* fall-through */
424 case 4:
425 break;
426 }
427 }
428 else {
429 /* Setup the VB pointers so that the next pipeline stages get
430 * their data from the right place (the program output arrays).
431 */
432 VB->ClipPtr = &store->results[VARYING_SLOT_POS];
433 VB->ClipPtr->size = 4;
434 VB->ClipPtr->count = VB->Count;
435 }
436
437 VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VARYING_SLOT_COL0];
438 VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VARYING_SLOT_COL1];
439 VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VARYING_SLOT_FOGC];
440 VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VARYING_SLOT_PSIZ];
441 VB->BackfaceColorPtr = &store->results[VARYING_SLOT_BFC0];
442 VB->BackfaceSecondaryColorPtr = &store->results[VARYING_SLOT_BFC1];
443
444 for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
445 VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
446 = &store->results[VARYING_SLOT_TEX0 + i];
447 }
448
449 for (i = 0; i < ctx->Const.MaxVarying; i++) {
450 if (program->info.outputs_written &
451 BITFIELD64_BIT(VARYING_SLOT_VAR0 + i)) {
452 /* Note: varying results get put into the generic attributes */
453 VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
454 = &store->results[VARYING_SLOT_VAR0 + i];
455 }
456 }
457
458
459 /* Perform NDC and cliptest operations:
460 */
461 return do_ndc_cliptest(ctx, store);
462 }
463
464
465 /**
466 * Called the first time stage->run is called. In effect, don't
467 * allocate data until the first time the stage is run.
468 */
469 static GLboolean
init_vp(struct gl_context * ctx,struct tnl_pipeline_stage * stage)470 init_vp(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
471 {
472 TNLcontext *tnl = TNL_CONTEXT(ctx);
473 struct vertex_buffer *VB = &(tnl->vb);
474 struct vp_stage_data *store;
475 const GLuint size = VB->Size;
476
477 stage->privatePtr = calloc(1, sizeof(*store));
478 store = VP_STAGE_DATA(stage);
479 if (!store)
480 return GL_FALSE;
481
482 /* a few other misc allocations */
483 _mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
484 store->clipmask = align_malloc(sizeof(GLubyte)*size, 32 );
485
486 return GL_TRUE;
487 }
488
489
490 /**
491 * Destructor for this pipeline stage.
492 */
493 static void
dtr(struct tnl_pipeline_stage * stage)494 dtr(struct tnl_pipeline_stage *stage)
495 {
496 struct vp_stage_data *store = VP_STAGE_DATA(stage);
497
498 if (store) {
499 GLuint i;
500
501 /* free the vertex program result arrays */
502 for (i = 0; i < VARYING_SLOT_MAX; i++)
503 _mesa_vector4f_free( &store->results[i] );
504
505 /* free misc arrays */
506 _mesa_vector4f_free( &store->ndcCoords );
507 align_free( store->clipmask );
508
509 free( store );
510 stage->privatePtr = NULL;
511 }
512 }
513
514
515 static void
validate_vp_stage(struct gl_context * ctx,struct tnl_pipeline_stage * stage)516 validate_vp_stage(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
517 {
518 if (ctx->VertexProgram._Current) {
519 _swrast_update_texture_samplers(ctx);
520 }
521 }
522
523
524
525 /**
526 * Public description of this pipeline stage.
527 */
528 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
529 {
530 "vertex-program",
531 NULL, /* private_data */
532 init_vp, /* create */
533 dtr, /* destroy */
534 validate_vp_stage, /* validate */
535 run_vp /* run -- initially set to ctr */
536 };
537